Defect recognition of tunnel pavement based on laser radar point cloud data and 3D reconstruction technology

The enclosed tunnel environment with restricted ventilation significantly hinders effective moisture dissipation, resulting in prolonged surface dampness. Under combined effects of vehicular loading and persistent humidity, exposed aggregates become prone to abrasion and detachment, substantially compromising pavement performance. This study develops an enhanced convolutional neural network (CNN) architecture that establishes a three-dimensional datacoupled model with self-adaptive capabilities for tunnel pavement defect recognition. Multi-scale fusion of subgrade morphological characteristics is used to enhance feature extraction capability. Laser point-cloud data integration optimizes sample equilibrium in optical datasets. The solution improves the accuracy of road microscopic defect detection. Experiments show that the 3D modeling effect of the improved convolutional network and the recognition accuracy of the machine vision detection speed are improved by 9.2% and 13.7%, respectively. The developed system exhibits exceptional detection accuracy in microscopic defect identification under complex tunnel conditions, as verified through field measurements. These technical advancements can provide valuable insights for infrastructure monitoring applications across multiple engineering domains.